LocalsOnlyevaluationsHow to read this
← Model shelfAnthropic

Claude Opus 4.8

Setups

Published full-model reference, then measured local recipes.

Recipe by benchmark scores. Published reference is not a rank. Blank cells have no cite.
RecipeGPQATerminalAutoVerified
Anthropic reference / publishedPublished reference———88.6

The published full-model reference is the top row — or the dashed gold tick on Graphic — not rank #1. A blank cell, missing tick, or missing bead has no recorded cite. Scores are never invented.

ON THIS SETUP · SOURCE-REPORTED

Anthropic reference

Cloud / API · Anthropic evaluation

1 benchmark result
Coding

SWE-bench Verified ↗

500 human-verified repository issues. Public-comparable headline for repo software engineering. Agent and harness belong on the cite.

Scoring & limits

swe-bench-verified · 500 tasks

This is the source’s reported aggregate. Its metric, agent configuration, and evaluation protocol determine how it should be interpreted. Read the original grading definition before comparing completion rates across setups.

88.6%500 evaluated tasks
Published · 2026-06-09Inspect evidence ↗

Task traces are not available for this setup.

Recipe, sources & reproducibilityExact artifacts and evaluation provenance
Runtime
Anthropic evaluation
Model host
Provider managed · physical host not recorded
Weight artifact
Provider managed
Weight revision
Not recorded

Context, sampling, harness settings and dataset versions can vary by benchmark. Each evidence link above contains its run-specific configuration.